The latest articles on DEV Community by gray sally (@graysally223). https://dev.to/graysally223 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4145296%2Ffb159b30-4f23-44cd-a921-4107084224fd.png https://dev.to/graysally223 en gray sally Mon, 28 Sep 2026 14:57:15 +0000 https://dev.to/graysally223/i-built-a-coremark-leaderboard-for-100-nas-cpus-heres-what-the-data-actually-shows-5fg2 https://dev.to/graysally223/i-built-a-coremark-leaderboard-for-100-nas-cpus-heres-what-the-data-actually-shows-5fg2 <p>Picking a CPU for a NAS or homelab server is mostly guesswork. Vendor spec sheets tell you core counts and boost clocks, but they don't tell you how an i3-12100T actually compares to a Ryzen 7 8840U for the kind of always-on, mixed workloads a home server runs. So I started collecting real CoreMark scores from real devices and put them on a public leaderboard: <a href="https://17nas.com" rel="noopener noreferrer">17NAS</a>.</p> <p>Every score is submitted by a device owner and human-reviewed before it lands on the chart. Here's what a few hundred submissions taught me.</p> <h2> The top isn't who you'd expect </h2> <p>The current leader is the AMD EPYC 7513 (32C/64T) at <strong>1,140,363</strong>, followed by the EPYC 7532 at 1,024,710. No surprise there — server silicon wins raw throughput. But scroll down and it gets interesting.</p> <h2> Efficiency is the real homelab story </h2> <p>Nobody runs a 280W EPYC in their closet. The interesting rows are the low-TDP parts:</p> <ul> <li> <strong>Ryzen 7 8840U</strong> (8C/16T, ~28W): 379,594 — beats a desktop i5-11500 (267,917) while sipping power</li> <li> <strong>i5-13500TE</strong> (35W): 436,649 — the "T tax" barely dents it</li> <li> <strong>i3-12100T</strong> (35W): 175,087 — for a pure file server, this is all you need</li> </ul> <p>If you're building today, the used-market sweet spot is last-gen mobile/handheld silicon, not old Xeons. A Xeon E5-2650 v4 still scores a respectable 584,273, but at 105W TDP and DDR4 platform costs, the efficiency math rarely works out anymore.</p> <h2> ARM is knocking </h2> <p>The <strong>Apple M4</strong> (ARMv9, 4P+6E) sits at 292,013 — roughly matching a desktop i9-13900H-class part in this benchmark. And the <strong>Snapdragon 8 Elite Gen 5</strong> shows up at 177,452. Neither is a NAS chip today, but the trajectory is obvious: ARM efficiency is coming for the low-power server segment, and it's worth watching.</p> <h2> Why CoreMark and not Geekbench? </h2> <p>CoreMark is small, deterministic, and runs on everything from a router SoC to an EPYC — which matters when your dataset spans x86, ARM, and whatever people have lying around. It's not a perfect proxy for NAS workloads (no I/O component), but as a pure CPU comparison across architectures, it's the most apples-to-apples number available.</p> <h2> The tools around it </h2> <p>The leaderboard is the core, but I also built two browser-based network tools that homelab folks keep asking for:</p> <ul> <li> <a href="https://17nas.com/ipv6-test.php" rel="noopener noreferrer">IPv6 connectivity test</a> — checks whether your network actually has working v6, not just what your ISP claims</li> <li> <a href="https://17nas.com/nat-test.php" rel="noopener noreferrer">NAT type detection</a> — figures out your NAT behavior, which matters if you're trying to reach your NAS remotely</li> </ul> <p>Both run entirely in the browser, no signup.</p> <h2> Run your own and submit </h2> <p>If you have a NAS, mini PC, or old server, <a href="https://17nas.com" rel="noopener noreferrer">run CoreMark and submit your score</a> — the chart gets more useful with every submission, especially for the oddball ARM boards and mini-PC chips that reviewers never benchmark.</p> homelab nas benchmarking cpu